Analysis of the impact of data granularity on privacy for the smart grid

Abstract

The upgrade of the electricity network to the "smart grid" has been intensified in the last years. The new automated devices being deployed gather large quantities of data that offer promises of a more resilient grid but also raise privacy concerns among customers and energy distributors.In this paper, we focus on the energy consumption traces that smart meters generate and especially on the risk of being able to identify individual customers given a large dataset of these traces. This is a question raised in the related literature and an important privacy research topic. We present an overview of the current research regarding privacy in the Advanced Metering Infrastructure. We make a formalization of the problem of de-anonymization by matching low-frequency and high-frequency smart metering datasets and we also build a threat model related to this problem. Finally, we investigate the characteristics of these datasets in order to make them more resilient to the de-anonymization process.Our methodology can be used by electricity companies to better understand the properties of their smart metering datasets and the conditions under which such datasets can be released to third parties.

Privacy preserving disclosure of authenticated energy usage data

“…Thus, in terms of privacy, it is not so significant. Moreover, such noisy data can also discourage re-identification attacks, such as the one discussed in [24], which attempts to find mappings between non-identifiable, high-frequency energy usage data and identifiable, low-frequency data.…”

A study on data de-pseudonymization in the smart grid

“…These types of data are called HighFrequency data (HF data) [4] and as mentioned earlier, raise privacy concerns. In earlier work [12] we show that even if the HF data is anonymized, an adversary can successfully link such an HF dataset with the corresponding LF dataset to identify a large portion of the households.…”

“…In the case of data from the AMI, previous research has shown that only a portion of the households' identities are protected, i.e. cannot be re-identified, as smart metering data characteristics can be used in the de-pseudonymization process [1,7,12]. However, it is important to better understand the conditions under which the suggested de-anonymization process works and how the number of protected households change given properties of the dataset obtained by the adversary.…”

How the quantity and quality of training data impacts re-identification of smart meter users?

“…Tudor et al [2] showed that when the two datasets (training and testing) contain consumption traces from the same period of time, re-identifying consumers is trivial by looking at the total electricity consumption of each trace. Most of the work in this space (including ours) assumes that training (the dataset with identifiers) and testing (the anonymized dataset) contain electricity consumption samples that are non-overlapping in time [3], [4], [5], [6].…”